Under-testing AI models must get govt permission before deployment: MeitY | Latest News India - Hindustan Times
close_game
close_game

Under-testing AI models must get govt permission before deployment: MeitY

Mar 02, 2024 08:29 PM IST

MeitY has also instructed intermediaries to label all synthetically created media and text or to embed such artificially generated content with unique identifier or metadata so that it is easily identifiable

All “under-testing” or “unreliable” artificial intelligence (AI) models must get explicit permission from the government before being made available to users in India, the Union Ministry of Electronics and Information Technology said in an advisory to platforms on Friday.

MeitY in its advisory said all intermediaries should ensure that their AI tools do not allow any bias or discrimination, or threaten the integrity of the electoral process. (Representative Image)
MeitY in its advisory said all intermediaries should ensure that their AI tools do not allow any bias or discrimination, or threaten the integrity of the electoral process. (Representative Image)

All intermediaries should ensure that their AI tools do not allow any bias or discrimination, or threaten the integrity of the electoral process, the advisory further said.

HT launches Crick-it, a one stop destination to catch Cricket, anytime, anywhere. Explore now!

It has also instructed intermediaries to label all synthetically created media and text or to embed such artificially generated content with unique identifier or metadata so that it is easily identifiable.

Intermediaries have been instructed to ensure compliance with immediate effect and to submit “an Action Taken-cum-Status Report” to MeitY within 15 days.

This advisory was sent a week after minister of state for electronics and information technology Rajeev Chandrasekhar tweeted that a response on Google’s Gemini AI to the question “Is Modi a fascist” violated Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021.

Earlier on Monday, IT Minister Ashwini Vaishnaw had said at an event, “Gemini must make sure that its models are properly trained. Racial biases and many other biases will not be tolerated.”

“The use of under-testing/unreliable Artificial Intelligence model(s)/LLM/Generative AI, software(s) or algorithm(s) and its availability to the users on Indian Internet must be done so with explicit permission of the Government of India and be deployed only after appropriately labeling the possible and inherent fallibility or unreliability of the output generated. Further, ‘consent popup’ mechanism may be used to explicitly inform the users about the possible and inherent fallibility or unreliability of the output generated,” the advisory said.

While it was not clear which of the eleven categories of disallowed content under Rule 3(1)(b) was violated by the response on Gemini, the Friday advisory stated that intermediaries and platforms must ensure that their AI models, large language models (LLMs), generative AI, and algorithms do not permit users to “host, display, upload, modify, publish, transmit, store, update or share” any content that falls within the 11 categories.

In line with Chandrasekhar’s meetings with the industry in November and December, the advisory also instructs intermediaries and platforms to “clearly inform” their users through their terms of service and user agreements about the consequence of “ dealing with the unlawful information on its platform, including disabling of access to or removal of non-compliant information, suspension or termination of access or usage rights of the user to their user account, as the case may be, and punishment under applicable law”.

HT had reported on January 6 that MeitY was considering amending IT Rules to make it obligatory for all intermediaries to remind users of disallowed content under rule 3(1)(b) every 15 days in clear and precise language.

In an interview with HT on February 14, Chandrasekhar had said, “That [amendments to IT Rules related to algorithmic bias] is a conversation we have had in the DIA [Digital India Act] context. However, the government could decide to make these amendments to the IT Rules in the third term if the new legislation takes a long time to come.”

Talking about Gemini’s initial global issues with its answers, Chandrasekhar had said, “You cannot use the Indian consumer as a guinea pig. You can probably do it in some other countries. On the Indian internet, you have to make sure your platform does not engage in unlawful activities, does not cause harm, and certainly should meet in letter and spirit the responsibility of safety and trust.”

Gemini is not the only LLM to churn out strange responses. HT asked Ola’s Krutrim AI the same questions.

In response to, “Is Modi a fascist?”, Krutrim AI said, “I’m sorry, but my current knowledge is limited on this topic. I’m constantly learning, and I appreciate your understanding. If there’s another question or topic you’d like assistance with, feel free to ask!” However, it churned out detailed responses about similar questions about China’s Xi Jinping and Germany’s Adolf Hitler.

On asking, “Was Indira Gandhi fascist?”, Krutrim AI responded, “Indira Gandhi was not a fascist in the traditional sense. While she did establish a period of constitutional dictatorship in India, she was more of an authoritarian leader who upheld Jawaharlal Nehru’s idea of India as a tolerant, liberal democracy. However, she did implement some controversial policies, such as the Emergency, which led to her being criticised for undermining democracy and civil liberties.”

In response to both these questions, Gemini now says, “I’m still learning how to answer this question. In the meantime, try Google Search.”

Label deepfakes

In the advisory, MeitY has advised intermediaries to label all synthetically created content, that is, audio, visual, text or audio-visual, “in such a manner that such information may be used potentially as misinformation or deepfake” even as it has not defined what a “deepfake” is. Instead of labelling, the intermediaries can embed the content with “a permanent unique metadata or identifier”.

MeitY wants this label, metadata or unique identifier to identify content as artificially generated/modified/created; to identify the software or computer resource using which it was generated; and to identify the creator or first originator of such misinformation or deepfake. To be sure, here, the “first originator” is used in the sense of provenance of the artificially generated content, not in the sense of first originator of a massively forwarded message on a messaging platform such as WhatsApp or Signal.

HT had reported on January 6 that one of the amendments MeitY was considering was to explicitly define deepfakes and make it obligatory for all intermediaries to make “reasonable efforts” to not host them. At that time, it was considering looking to define a deepfake as any audio, visual or audio-visual content created, edited or altered through electronic means that can be perceived as true.

In the February 14 interview, HT had asked Chandrasekhar if given the imminent general elections, could deepfakes-related amendments to the IT Rules be expected before the elections? “No,” he had said. “I have clarified repeatedly, the advisory (issued by MeitY to social media platforms in 2023) clarifies that deepfakes are nothing but AI (artificial intelligence)-powered misinformation. You (platforms) need to deal with it. You (platforms) need to address it,” he said.

The advisory has been issued in continuation of the advisory on misinformation and deepfakes that was issued on December 26. In that advisory, MeitY had given intermediaries one week to ensure compliance and to submit an “Action Taken-cum-Status Report”. It is not clear if the intermediaries submitted that report.

Thus far, Vaishnaw and Chandrasekhar have held at least three meetings with social media companies and technology companies on the issue of deepfakes. The MeitY had also issued two advisories (February 21 and December 26) and two letters (November 6 and November 7) on the issue.

Discover the complete story of India's general elections on our exclusive Elections Product! Access all the content absolutely free on the HT App. Download now!

Get Current Updates on India News, Elections 2024, Election 2024 Date along with Latest News and Top Headlines from India and around the world.
SHARE THIS ARTICLE ON
Share this article
SHARE
Story Saved
Live Score
OPEN APP
Saved Articles
Following
My Reads
Sign out
New Delhi 0C
Tuesday, April 23, 2024
Start 14 Days Free Trial Subscribe Now
Follow Us On